In one workshop day you design, test and improve an AI agent with Microduck that doesn’t just talk, but also acts: choosing, setting limits and checking whether it really worked.
The scenario: Microduck is a robot assistant in a demonstration room. It stands in a practice area and knows a few tested actions. There are no rules yet. You are the member of staff.
What you practise with Microduck applies to every agent that does something: choosing from permitted actions, enforcing boundaries and checking whether it really worked.
Simulator · Microduck’s practice area
Can you get Microduck to score 8 out of 8?
Microduck is a robot assistant in a demonstration room. It knows five tested actions, but has no rules yet. Play test cards, find out where it goes wrong and choose the rule that fixes it. This is exactly what you do on the course, but with a real Microduck in a practice area.
Ready for the first test card
1
Do something fun!
Play a test card
Each test card is an instruction from the course. You can also type your own instruction. Microduck chooses what to do and carries it out in the practice area.
2
WrongCould be betterGood
Watch what Microduck does
It works with the rules it has. At the start, that is zero. Red is wrong, orange could be better, green is good.
3
✓ Ask when in doubtInstruction or software?
Choose the rule, and where it belongs
Did it go wrong? Choose which rule would have prevented it, and sometimes where it belongs: in the instruction or hard-coded in the software. Passed so far: 0/8
Robot assistantRules0/7Tests0/8
Emergency stopWorks directly on the robot, bypassing the AI. A real robot should always have one.
0/5 stars0 points0:00
Click the floor to drive←↑→ steerspace roller-skateTap the floor or use the joystick
Driving game
Collect the 5 stars in the practice area.
Click the floor and Microduck drives there.
Tap the floor and Microduck drives there. Or use the joystick.
Or steer yourself: ←→ turn, ↑ drive, space roller-skate.
Here you are the controller: there is no AI in between. The only thing the software guards is the practice area, if that rule is switched on.
Ready to test
Microduck is standing in the practice area. It has no rules yet. Play the first test card and see what happens.
The 3D view can’t load on this device. The test cards, the rules and the report still work.
Under the hood · what happens between instruction and movement?
1Instruction
2Agent chooses
3Control layer
4Controller executes
SentReported backVerified
The agent chooses an action, the controller carries it out. In between sits the control layer, which enforces what is and isn’t allowed. And only when someone has checked the result do you know it really worked.
Who decided here? · module 1 in miniature
Microduck waves to the visitor three times. Watch them one by one and predict: who or what decided that it would wave?
Who decided that Microduck would wave?
The practice material
Microduck is the means, not the goal.
On this course you learn to work with AI agents that act. To make that visible, you practise with Microduck: a small two-legged robot duck from Pollen Robotics and Hugging Face. What you learn about instructions, boundaries and checks, you apply to every agent that does something: in software, in your processes or on a robot.
±25cm
Small and light: safe to practise with in a marked-out practice area.
15
Motors in its legs, neck and head. Every movement is an action an agent can choose.
3–5
Tested actions on the course. The depth comes from better instructions, not from more tricks.
2
Microducks during the course: one per team. They stay ours; you don’t take them home.
Microduck is a product of Pollen Robotics and Hugging Face, with open-source software. It waddles, roller-skates, gets back up by itself and picks up small things with its beak. Smairtly uses it as practice material, because with a robot you immediately see what an agent chooses and does.
How Microduck is built
Click on a partand see what it does, and what it can’t do.
Who is it for?
For anyone who chooses, designs or introduces AI agents
Employees, managers, process owners and innovation professionals who want to understand what AI agents can do, where the limits are and how to use them responsibly.
Agent, automation or robot?
You learn when a fixed automation is enough, when an agent adds value, and when a robot does.
Boundaries that really work
You discover which boundary belongs in the instruction, and which the software itself must enforce.
One Microduck per team
During the course, each team of three to four works with its own Microduck. Hands-on work comes first.
The programme
Eight modules. One day.
Follow Microduck through the day. Each module combines a short explanation with a hands-on exercise. Click a module to see its content and what you take away.
Three demonstrations with Microduck: a button, a fixed sequence and a text instruction
For each demonstration: who decided what happened, what was fixed in advance and what choice did the system make?
Why movement alone says nothing about how smart or autonomous a robot is
You take away: a clear picture of what a robot really decides by itself.
Chatbot, fixed automation, AI agent and robot with an agent side by side
Two layers: the agent chooses an action, the controller carries out the movement
Breaking down an instruction: goal, information, actions, boundaries and when it has worked
You take away: your first design of a simple robot task.
The five building blocks: goal, available actions, boundaries, an agreement on ambiguity and a recognisable end point
Skills you can use straight away with any AI agent that actually does something: on a robot, in your software or both.
Understand the difference between a chatbot, a fixed automation and an AI agent.
Know what changes when an agent’s choices have physical consequences.
Give an agent a clear instruction: a goal, the information it needs, permitted actions and boundaries.
Know when an agent should ask a follow-up question, stop or call in a person.
Know which boundaries belong in the instruction and which the software must enforce.
Check whether something really worked, instead of trusting ‘done!’.
Test a robot application for mistakes and exceptions, and record what happened.
Translate it into a first application and a small trial in your own organisation.
Official learning outcomes · handy for your training budget
Afterwards, the participant can:
Explain the difference between a chatbot, a fixed automation and an agent.
Translate a goal into instructions, information and permitted actions.
Decide when an agent should ask a follow-up question or call in a person.
Test a robot application for mistakes and exceptions.
Develop a first application for their own organisation.
Practical information
Everything you need to know.
Duration & location
One workshop day on a Tuesday or Thursday in Utrecht, Amsterdam, Rotterdam, Eindhoven or Arnhem. Or in-company at your location. Groups of 3 to 8 participants, guaranteed to run from 3.
Price
Introductory price€545€695
Only for the first two sessions: you save €150. After that €695 per person. Prices exclude VAT.
In-company: price tailored to your team
Included
Two Microducks during the course (one per team)
Coaching, laptops and worksheets
Lunch for open enrolment
Afterwards
Your test report, use-case card, personal action plan and a certificate of participation. The Microducks stay with us.
Your trainer
Who runs the course?
Marcel Boeren
Founder of Smairtly · AI strategist, trainer and speaker
Through Smairtly, Marcel helps organisations not only to understand AI, but also to apply it in practice. From AI adoption and strategy to developing smart agents and automations.
On this course Marcel takes you to the next step: Physical AI. What happens when AI doesn’t just process information, but can also perceive, choose and control movement? With practical exercises and robots you discover how AI, software and the physical world come together, and what possibilities that opens up for the future of work.
Frequently asked questions
Good to know.
Do I need to be able to programme?
No. You work with input fields, action buttons and adjustable rules in a prepared environment. No programming knowledge or robot experience is needed.
I have never used ChatGPT or Copilot. Can I take part?
Yes. Some experience is helpful, but the first modules start with a short introduction, so everyone can take part.
How many participants are there, and do I work with the robot myself?
We work with six to eight participants in two teams of three to four. During the course there are two Microducks: one per team, so you get hands-on yourself. Within your team you switch roles: client, agent designer and tester.
Does the course always go ahead?
A date is guaranteed to run as soon as there are 3 participants: then it will definitely go ahead. If a date doesn’t reach that, we’ll contact you in good time to choose another date together. There are a maximum of 8 places per group.
Is the course mainly about Microduck?
No. It’s about the technique: how to give an AI agent a clear instruction, which boundaries you enforce in the software and how to check whether something really worked. Microduck is the practice material. With a robot you immediately see what an agent chooses and does, which makes it easy to talk about. What you learn, you also use with agents that have no robot.
What is the difference from the Reachy Mini course?
Reachy Mini is all about conversation: what an AI assistant says, when it asks follow-up questions and when it hands over to a person. Microduck is all about action: which action an agent chooses, which boundaries the software enforces and how you check whether something really worked. The courses complement each other and you can take them separately. See the Reachy Mini course.
Is it safe? What if the robot does something it shouldn’t?
Microduck is small and light and works in a marked-out practice area. We only use movements that have been tested in advance, the software limits what it may do and the trainer keeps watch over the area. And there is always a stop function that works immediately, bypassing the AI.
How much does the course cost?
The first two sessions are introductory days at €545 per person. After that the price is €695 per person. Prices exclude VAT. Coaching, laptops, the training environment, all worksheets and (for open enrolment) lunch are included.
Where does the course take place?
In Utrecht, Amsterdam, Rotterdam, Eindhoven or Arnhem, on Tuesdays and Thursdays. You’ll receive the exact address with the confirmation of your booking. At your location is also possible: then we bring the robots, laptops and practice material to you.
What do I need to bring?
Just a process or task from your own work. We’ll provide the laptops and the robots.
Can I use a case from my own organisation?
For open enrolment we work with fictional cases, so nobody has to bring confidential information. In the final modules you translate the lessons to your own work. For an in-company course we can use your own processes.
What happens with the camera and microphone?
We work with fictional practice situations without personal data. Beforehand we explain whether and where images and sound are processed.
What do I take home?
Your agent design, your tested instruction and rules, your test report, a use-case card and a personal action plan. The prototype works in the training environment. The Microducks remain Smairtly’s: you don’t take a robot home, and ongoing access to accounts is not automatically included.
Can we also organise the course for our own team?
Yes. Choose ‘At your location’ when you sign up, and we’ll agree the date, group size and content together.
Will I receive a certificate of participation?
Yes. Afterwards you receive a certificate of participation from Smairtly, along with your test report, use-case card and action plan.
Can I cancel or reschedule?
Contact us at [email protected] and we’ll find a solution together, such as another date or a colleague in your place. You’ll receive the terms with the confirmation of your booking.